How to improve patient safety by text mining with medical incident reports: Innovative technologies using e-health and health technology assessment

We propose a new approach to detect the precarious situation in medical care and solve the communication-gap by analyzing tracking record. We evaluated the degree of similarities between incident documents obtained bottom-up and the links between existing classes granted top-down. We made it possible to evaluate overall similarities regarding incident documents with the techniques of natural language processing and network analysis with more than 20,000 reports. In this research, we evaluated the degree of similarities between incident documents obtained bottom-up and the links between existing classes granted top-down. We made it possible to evaluate overall similarities regarding incident documents by using the method of network analysis. With regard to the background, the results of the analysis demonstrated that compared with abstract and solution, existing classes are inadequate for representing the characteristics of documents and that there is a need to improve classes. Some categorizes by top-down analysis don't reflect the category by the bottom-up analysis. Our results suggest the effectiveness of introducing the network analysis method. We made it possible to analyze the differences of understanding of the incident reports between doctors and nurses. We attempted the consensus building that depended bottom-up by the network analysis.